Magnetic resonance imaging apparatus and method of reconstructing MR image by using neural network

Abstract

A magnetic resonance imaging (MRI) apparatus includes a processor, and a memory storing a program including instructions that, when executed by the processor, cause the processor to acquire first data of a subsampled magnetic resonance (MR) image, acquire, based on a learning model using a neural network, first reconstructed data with respect to rows of pixels in a first phase encoding direction of the first data of the subsampled MR image, and obtain a reconstructed image corresponding to the subsampled MR image, using the first reconstructed data.

What is claimed is: 
     
       1. A magnetic resonance imaging (MRI) apparatus comprising:
 a processor; and 
 a memory storing at least one program comprising instructions that, when executed by the processor, cause the processor to:
 acquire data of at least one subsampled magnetic resonance (MR) image and data of at least one fully sampled MR image; 
 obtain a learning model using at least one neural network by learning correlations between the at least one subsampled MR image and the at least one fully sampled MR image in units of at least one row of pixels corresponding to a phase encoding direction; 
 acquire data of a subsampled MR image; 
 divide the data of the subsampled MR image into a plurality of groups, each corresponding to the at least one row of pixels corresponding to the phase encoding direction; 
 acquire, based on the learning model using the at least one neural network, reconstructed data respectively corresponding to the plurality of groups with respect to the at least one row of pixels corresponding to the phase encoding direction of the data of the subsampled MR image; and 
 obtain a reconstructed image corresponding to the subsampled MR image, using the acquired reconstructed data. 
 
 
     
     
       2. The MRI apparatus of  claim 1 , wherein the memory stores the at least one program comprising the instructions that, when executed by the processor, further cause the processor to:
 acquire subsampled real image data and subsampled imaginary image data; 
 divide the subsampled real image data and the subsampled imaginary image data into the plurality of groups, each corresponding to the at least one row of pixels corresponding to the phase encoding direction. 
 
     
     
       3. The MRI apparatus of  claim 1 , wherein the memory stores the at least one program comprising the instructions that, when executed by the processor, further cause the processor to:
 acquire subsampled magnitude image data and subsampled phase image data; 
 divide the subsampled magnitude image data and the subsampled phase image data into the plurality of groups, each corresponding to the at least one row of pixels corresponding to the phase encoding direction. 
 
     
     
       4. The MRI apparatus of  claim 1 , wherein the memory stores the at least one program comprising the instructions that, when executed by the processor, further cause the processor to dynamically determine the plurality of groups into which the data of the subsampled MR image is divided, using the at least one neural network. 
     
     
       5. The MRI apparatus of  claim 1 , wherein the memory stores the at least one program comprising the instructions that, when executed by the processor, further cause the processor to:
 acquire data of at least one additional image by circularly shifting the data of the subsampled MR image in a direction corresponding to the phase encoding direction; and 
 respectively divide the data of the at least one additional image into the plurality of groups, each corresponding to the at least one row of pixels corresponding to the phase encoding direction. 
 
     
     
       6. The MRI apparatus of  claim 1 , wherein the subsampled MR image comprises a subsampled three-dimensional (3D) MR image including a first phase encoding direction and a second phase encoding direction, and
 wherein the memory stores the at least one program comprising the instructions that, when executed by the processor, further cause the processor to:
 divide the data of the subsampled 3D MR image into a plurality of first groups, each corresponding to at least one row of pixels corresponding to the first phase encoding direction; and 
 acquire, based on the learning model using the at least one neural network, first reconstructed data respectively corresponding to the plurality of first groups. 
 
 
     
     
       7. The MRI apparatus of  claim 6 , wherein the memory stores the at least one program comprising the instructions that, when executed by the processor, further cause the processor to:
 divide the first reconstructed data respectively corresponding to the plurality of first groups into a plurality of second groups, each corresponding to at least one row of pixels corresponding to the second phase encoding direction; 
 acquire second reconstructed data respectively corresponding to the plurality of second groups based on the learning model using the at least one neural network; and 
 obtain the reconstructed image corresponding to the subsampled 3D MR image by using the acquired second reconstructed data. 
 
     
     
       8. The MRI apparatus of  claim 1 , further comprising a multi-channel radio frequency (RF) coil configured to receive MR signals from an object,
 wherein the memory stores the at least one program comprising the instructions that, when executed by the processor, further cause the processor to:
 acquire the data of the subsampled MR image corresponding to the multi-channel RF coil; and 
 obtain the reconstructed image, based on a sensitivity of the multi-channel RF coil and the reconstructed data. 
 
 
     
     
       9. The MRI apparatus of  claim 1 , further comprising a multi-channel radio frequency (RF) coil configured to receive MR signals from an object,
 wherein the memory stores the at least one program comprising the instructions that, when executed by the processor, further cause the processor to acquire, based on the MR signals received via the multi-channel RF coil, the data of the subsampled MR image obtained by subsampling in a regular or irregular pattern along the phase encoding direction. 
 
     
     
       10. An image reconstruction method comprising:
 acquiring data of at least one subsampled magnetic resonance (MR) image and data of at least one fully sampled MR image; 
 obtaining a learning model using at least one neural network by learning correlations between the at least one subsampled MR image and the at least one fully sampled MR image in units of at least one row of pixels corresponding to a phase encoding direction; 
 acquiring data of a subsampled MR image; 
 dividing the data of the subsampled MR image into a plurality of groups, each corresponding to the at least one row of pixels corresponding to the phase encoding direction; 
 acquiring, based on the learning model using the at least one neural network, reconstructed data respectively corresponding to the plurality of groups with respect to the at least one row of pixels corresponding to the phase encoding direction of the data of the subsampled MR image; and 
 obtaining a reconstructed image corresponding to the subsampled MR image, by using the acquired reconstructed data. 
 
     
     
       11. The image reconstruction method of  claim 10 , wherein the acquiring of the data of the subsampled MR image comprises respectively acquiring subsampled real image data and subsampled imaginary image data, and
 wherein the dividing of the data of the subsampled MR image into the plurality of groups comprises respectively dividing the subsampled real image data and the subsampled imaginary image data into the plurality of groups, each corresponding to the at least one row of pixels corresponding to the phase encoding direction. 
 
     
     
       12. The image reconstruction method of  claim 10 , wherein the acquiring of the data of the subsampled MR image comprises respectively acquiring subsampled magnitude image data and subsampled phase image data, and
 wherein the dividing of the data of the subsampled MR image into the plurality of groups comprises respectively dividing the subsampled magnitude image data and the subsampled phase image data into the plurality of groups, each corresponding to the at least one row of pixels corresponding to the phase encoding direction. 
 
     
     
       13. The image reconstruction method of  claim 10 , wherein the acquiring of the data of the subsampled MR image comprises acquiring data of at least one additional image by circularly shifting the data of the subsampled MR image in a direction corresponding to the phase encoding direction, and
 wherein the dividing of the data of the subsampled MR image into the plurality of groups comprises respectively dividing the data of the subsampled MR image and the data of the at least one additional image into the plurality of groups, each corresponding to the at least one row of pixels corresponding to the phase encoding direction. 
 
     
     
       14. The image reconstruction method of  claim 10 , wherein the subsampled MR image comprises a subsampled three-dimensional (3D) MR image including a first phase encoding direction and a second phase encoding direction,
 wherein the dividing of the data of the subsampled MR image into the plurality of groups comprises dividing the data of the subsampled 3D MR image into a plurality of first groups, each corresponding to at least one row of pixels corresponding to the first phase encoding direction, and 
 wherein the acquiring of the reconstructed data comprises acquiring first reconstructed data respectively corresponding to the plurality of first groups based on the learning model using the at least one neural network. 
 
     
     
       15. The image reconstruction method of  claim 14 , further comprising:
 dividing the first reconstructed data respectively corresponding to the plurality of first groups into a plurality of second groups, each corresponding to at least one row of pixels corresponding to the second phase encoding direction, and 
 acquiring second reconstructed data respectively corresponding to the plurality of second groups based on the learning model using the at least one neural network, 
 wherein the obtaining of the reconstructed image comprises obtaining the reconstructed image corresponding to the subsampled 3D MR image by using the acquired second reconstructed data. 
 
     
     
       16. The image reconstruction method of  claim 10 , wherein the acquiring of the data of the subsampled MR image comprises acquiring the data of subsampled MR images corresponding to a multi-channel radio frequency (RF) coil, and wherein the obtaining of the reconstructed image corresponding to the subsampled MR image comprises obtaining the reconstructed image based on sensitivity of the multi-channel RF coil and the reconstructed data. 
     
     
       17. The image reconstruction method of  claim 10 , further comprising receiving MR signals from an object via a multi-channel radio frequency (RF) coil,
 wherein the acquiring of the data of the subsampled MR image comprises acquiring, based on the MR signals received via the multi-channel RF coil, the data of the subsampled MR image obtained by subsampling in a regular or irregular pattern along the phase encoding direction. 
 
     
     
       18. A computer program product comprising a computer-readable recording medium having recorded thereon a program for executing the image reconstruction method of  claim 10  on a computer. 
     
     
       19. A magnetic resonance imaging (MRI) apparatus comprising:
 a processor; and 
 a memory storing at least one program, 
 wherein the memory stores, when the at least one program is executed by the processor, instructions that cause the processor to perform:
 acquiring data of at least one subsampled magnetic resonance (MR) image and data of at least one fully sampled MR image; 
 obtaining a learning model using at least one neural network by learning correlations between the at least one subsampled MR image and the at least one fully sampled MR image in units of at least one row of pixels corresponding to an aliasing direction in which aliasing occurs in the data of the subsampled MR images; 
 acquiring data of a subsampled MR image; 
 dividing the data of the subsampled MR image into a plurality of groups, each corresponding to the at least one row of pixels corresponding to the aliasing direction; 
 acquiring, based on the learning model using the at least one neural network, reconstructed data respectively corresponding to the plurality of groups; and 
 obtaining a reconstructed image corresponding to the subsampled MR image, by using the acquired reconstructed data. 
 
 
     
     
       20. An image reconstruction method comprising:
 acquiring data of at least one subsampled magnetic resonance (MR) image and data of at least one fully sampled MR image; 
 obtaining a learning model using at least one neural network by learning correlations between the at least one subsampled MR image and the at least one fully sampled MR image in units of at least one row of pixels corresponding to an aliasing direction in which aliasing occurs in the data of the subsampled MR image; 
 acquiring data of a subsampled MR image; 
 dividing the data of the subsampled MR image into a plurality of groups, each corresponding to the at least one row of pixels corresponding to the aliasing direction; 
 acquiring, based on the learning model using the at least one neural network, reconstructed data respectively corresponding to the plurality of groups; and 
 obtaining a reconstructed image corresponding to the subsampled MR image, by using the acquired reconstructed data.